Does every number fall to one?

Take any positive whole number. If it is even, halve it; if it is odd, triple it and add one. Repeat. Start with 7 and you get 22, 11, 34, 17, 52, 26, 13, 40, 20, 10, 5, 16, 8, 4, 2, 1. The conjecture, posed by Lothar Collatz in 1937, is that every starting number reaches 1. Every number ever tested does, and there are a great many: everything up to roughly three hundred quintillion. In mathematics that counts as suggestive scenery, not proof.

Why it matters

Honestly: not for applications. The Collatz conjecture is in this catalog because it is the purest specimen of a humbling species, the problem trivial to state and apparently impossible to prove. Erdős said mathematics is not yet ripe for such questions, and offered five hundred dollars for a proof anyway. Its difficulty is itself the finding. The iteration mixes two incompatible arithmetic worlds, halving and tripling, and mathematics currently owns no tools for predicting the long-run behavior of even this simplest such hybrid. Whatever finally cracks it will likely be a genuinely new idea, and new ideas about the interface of order and pseudo-randomness in arithmetic would echo far beyond one puzzle.

What has been tried

Brute force has verified the conjecture to astronomical heights, and statistical heuristics explain why it ought to be true: an average step shrinks a number, since halving happens more often than tripling wins back. Rigorous results circle the target. Almost all numbers eventually dip below their starting value, and in 2019 Terence Tao proved that almost all Collatz orbits sink to almost bounded values, about as close as current methods can squeeze without resolving the question. From the other side comes a warning: John Conway showed that for natural generalizations of Collatz-style rules, the halting question is formally undecidable, meaning no algorithm can settle every such system. The original conjecture might, in principle, be unprovable rather than merely unproven.

Where the edge is

The gap between almost all numbers and all numbers is the entire problem. One divergent orbit, or one hidden loop other than the familiar 4, 2, 1, would falsify the conjecture, and nothing known forbids either.

What would count as an answer

A proof that every orbit reaches 1, a counterexample that never does, or a proof that the question cannot be settled from standard axioms. Any of the three would be a landmark. The smart money refuses to say which.